with whom Judges FARRIS, FERNANDEZ, T.G. NELSON and KLEINFELD join, dissenting.
What makes this a tough case is that Robert Dunbar, the prosecution’s key witness, turns out to have been a big liar. Had the jurors seen Dunbar as we now see him, they would not have believed a thing he said. It is therefore surprising that my colleagues in the majority are able to say with confidence that one of Dunbar’s many stories accurately describes what happened the day Robert Shaw was killed. And yet, it is this finding-that Dunbar told the truth when he said that he and his wife killed Shaw-which unlocks the Schlup gateway through which Carriger must pass before we can consider his abusive Brady claim.'
Dunbar’s fairy tale is not the stuff of which Schlup claims are made. The Supreme Court recognized in Schlup that “a substantial .claim that constitutional error has caused the conviction of an innocent person is extremely rare.” Schlup v. Delo, 513 U.S. 298, 324, 115 S.Ct. 851, 130 L.Ed.2d 808 (1995). To make out such a claim, a petitioner must come forward with reliable new evidence of his innocence. A repudiated recantation can never be reliable enough to satisfy Schlup. But even if it could under ideal circumstances, it cannot here because the majority’s new evidence, even if believed, proves merely that Dunbar was guilty of Shaw’s murder right along with Carriger. Only the innocent can profit from Schlup, and the new evidence here leaves Carriger at the murder scene with blood on his hands. Because the majority widens the Schlup gateway far beyond what the Supreme Court envisioned, I must respectfully dissent.
*483I
Schlup is a narrow exception to' the rule that we will not consider abusive habeas claims; its limited purpose is to protect against “the execution of a person who is entirely innocent.” 513 U.S. at 324-25, 115 S.Ct. at 866 (emphasis added). As a matter of necessity and design, we normally rely on a state’s trial, appellate and collateral review processes to determine the truth in criminal prosecutions. Federal habeas provides another layer of protection, albeit limited. In capital cases, Schlup provides a further layer, more limited still. Only where a petitioner makes a compelling case based on reliable, credible, concrete proof that he is actually innocent may a federal court cast aside orderly procedure and consider an otherwise defaulted habeas claim.
Schlup started by recognizing that the vast majority of innocence claims are not credible: “[Challenges to the propriety of imposing a sentence of death are routinely asserted in capital eases_” Id. at 324, 115 S.Ct. at 865. Because unreliable evidence of innocence is easy to come by, the district court should quickly dismiss claims based on such evidence. Id. Truly credible claims are far less common: “[Experience has taught us that a substantial claim that constitutional error has caused the conviction of an innocent person is extremely rare.” Id. Only when a petitioner comes up with reliable, concrete, verifiable proof of actual innocence do we subordinate principles of finality and comity by considering a proeedurally defaulted claim. Id.
Writing for the Court in Schlup, Justice Stevens explained that “[t]o be credible, [a claim of actual innocence] requires petitioner to support his allegations of constitutional error with new reliable evidence ... that was not presented at trial.” Id. at 324, 115 S.Ct. at 865. Requiring that new-evidence be reliable draws a workable line between the many cases involving frivolous claims of innocence and the few that raise potentially meritorious claims. See Weeks v. Bowersox, 119 F.3d 1342,1351 (8th Cir.1997) (en banc); Bowman v. Gammon, 85 F.3d 1339, 1346 (8th Cir.1996). Schlup offered three examples of evidence that would pass the threshold of reliability: exculpatory scientific evidence, trustworthy eyewitness accounts and critical physical evidence. Schlup, 513 U.S. at 324, 115 S.Ct. at 865-66. By enumerating the kinds of proof that could form the basis of a substantial claim of innocence, the Supreme Court was telling us that less reliable kinds of evidence cannot support an actual innocence claim. The Court clearly did not hold that Schlup may be invoked whenever a witness changes his story.
Carriger’s new evidence is not exculpatory scientific evidence, arid it is not critical physical evidence. Is it a “trustworthy eyewitness account”? Certainly not. Recanting testimony'has long been disfavored as the basis for a claim of innocence. Appellate courts, even on direct review, look upon recantations with extreme suspicion. See United States v. Pointer, 17 F.3d 1070, 1074 (7th Cir.1994); Olson v. United States, 989 F.2d 229, 231 (7th Cir.1993); United States v. Nixon, 881 F.2d 1305, 1311 (5th Cir.1989); United States v. Santiago, 837 F.2d 1545, 1550 (11th Cir.1988); United States v. DiPaolo, 835 F.2d 46, 49 (2d Cir.1987); United States v. Adi, 759 F.2d 404, 408 (5th Cir.1985).1 Recanting testimony is easy to find but difficult to confirm or refute: witnesses forget, witnesses disappear, witnesses with personal motives change their stories many times, before and after trial. A repudiated recantation like Dunbar’s is even less worthy of serious consideration. One respected commentator advises that a motion for a new trial based on a repudiated recantation usually isn’t worth the time it takes to prepare: ’ “Although [a repudiation] does suggest that the witness is ‘completely, irresponsible,’ the motion appears to have little or no chance of success. With the repudiation of the recantation, it *484becomes merely impeaching and could be used at a new trial only for the purpose of cross-examining the witness, and not as substantive evidence.” Moore’s Federal Practice § -633.05[2] (3d ed.1997).
The majority tries to add force to Dunbar’s recantation with the high drama of a death bed confession, the mea culpa of a guilt-ridden man clearing his conscience before going to meet his Maker.2 See maj. op. at 467. Quite moving, were it but true. Dunbar long outlived his confession: He recanted the confession shortly after he made it, but didn’t die until four years later. Dunbar’s recantation was nothing but a spiteful coun-terpunch to his estranged wife’s accusation. In post-conviction proceedings nine years after the murder, the defense had called Joyce Stevens, Dunbar’s ex-wife, who had provided Dunbar’s “nap” alibi at trial. Joyce’s new story was that Dunbar had confessed he was at the jewelry store with Carriger, and that Dunbar had actually struck the fatal blows with the skillet.3 When Dunbar took the stand a few weeks later, he parried Joyce’s story by testifying that she robbed the place with him and struck the first blows with the frying pan. Tit for tat.
The majority makes much of the fact that Dunbar confessed in “open court.” See maj. op. at 465 (“Dunbar actually confessed in open court”); id. at 465 (“Dunbar confessed under oath in open court”). But Dunbar also testified at trial in open court, and then repudiated -his confession in open court six weeks after he had made it. We thus know for sure that Dunbar lied in court at least once, probably many more times. See id. at 481. The majority makes Dunbar out to be a congenital liar, see id. at 480-482 then treats one of his statements as the Gospel Truth.
As in many cases, recanting testimony is all too common here: Joyce recanted her trial testimony, her children recanted theirs, Dunbar recanted his, and then he recanted his recantation. It’s not easy to pluck the truth from this thistle of lies. The majority credits both Joyce’s and Dunbar’s recantations, but the two stories contradict each other on the key fact of whether Carriger was there. Far from corroborating Dunbar’s confession, Joyce’s testimony yanks the rug out from under the one fact that matters. To unlock the Schlwp gateway for Carriger, we must find that Dunbar’s recanted confession-not his testimony at trial, not his retraction, not Joyce’s testimony at trial or her retraction-accurately describes what happened. A confession shrouded by so much doubt from a witness who is now dead is not reliable enough to satisfy Schlup. If what we have here is enough, the Schlup gateway becomes a freeway off death row. The doctrine of abuse of the writ, see McCleskey v. Zant, 499 U.S. 467, 493, 111 S.Ct. 1454, 1469-70, 113 L.Ed.2d 517 (1991), will be severely undermined.
II
The Schlup framework protects only those who are “entirely” innocent, not those who are guilty of homicide but ineligible for the death penalty. Under Schlup, “the petitioner must show that it is more likely than not that no reasonable juror would have convicted him in the light of the new evidence.” Schlup v. Delo, 513 U.S. 298, 327, 115 S.Ct. 851, 867, 130 L.Ed.2d 808 (1995). This compact phrase formulates a daunting standard: We must imagine twelve reasonable jurors hearing the original case against the petitioner, as augmented by the new evidence. For a petitioner to pass through the Schlup gateway, he must persuade us that every imaginary juror, twelve out of twelve, would vote *485to acquit him of any involvement in the killing.4
In a short paragraph buried in the bowels of its opinion, the majority dismisses the notion that “Carriger must prove that he is actually innocent.” Maj. op. at 478. According to the majority, under Schlup “the petitioner is required to present ‘evidence of innocence’ such that ‘a court cannot have confidence in the outcome of the trial.’ ” Id. The majority cuts the heart out of Schlup by omitting its key requirement, namely that “a petitioner show that he is ‘actually innocent.’ ” Schlup, 513 U.S. at 327, 115 S.Ct. at 867. The showing of innocence under Schlup /Carrier is somewhat lower than that under Herrera/Sawyer5 but the need for a petitioner to proye his innocence is echoed repeatedly in Schlup: “Carrier requires a petitioner to show that he is ‘actually innocent.’ ” Schlup, 513 U.S. at 327, 115 S.Ct. at 867; id. at 326-27, 115 S.Ct. at 867 (“[W]e hold that the Carrier “probably resulted” standard ... must govern the miscarriage of justice inquiry when a petitioner who has been sentenced to death raises a claim of actual innocence to avoid a procedural bar to the consideration of the merits of his constitutional claims.”); id. at 327, 115 S.Ct. at 867 (“The Carrier standard requires the habeas petitioner to show that ‘a constitutional violation has probably resulted in the conviction of one who is actually innocent.”); id. at 327, 115 S.Ct. at 867 (“The Carrier standard is intended to focus the inquiry on actual innocence.”); id. at 327, 115 S.Ct. at 867 (“[T]he Carrier standard requires a petitioner to show that it is more likely than not that ‘no reasonable juror’ would have convicted him.”).6
The majority’s disregard of Schlup shows up in another stealth reference, this time in the bottom half of a footnote: “We do not agree with the assumption that ... our test under Schlup is to decide how a hypothetical *486jury would regard each bit of new evidence. Our task is to determine whether confidence in the actual verdict is undermined.” Maj. op. at 474 n. 4. Are we all reading the same Supreme Court opinion? In my version, the Court holds that “a petitioner does not meet the threshold requirement unless he persuades the district court that, in light of the new evidence, no juror, acting reasonably, would have voted to find him guilty beyond a reasonable doubt.” Schlup, 513 U.S. at 329, 115 S.Ct. at 868 (emphasis added).7 The Supreme Court expects us to consider the state of the evidence now-adding both the exculpatory and the inculpatory proof that has come to light since the verdict-and determine whether a properly instructed jury would find the defendant innocent. Predicting what reasonable jurors would do is not easy.8 It’s much harder than combing the record and picking at the state’s case. But it’s what Schlup demands.
Ill
Carriger has failed to prove that, “in light of the new evidence, no juror, acting reasonably, would have voted to find him guilty beyond a reasonable doubt.” Schlup, 513 U.S. at 329, 115 S.Ct. at 868. Even if Dunbar’s repudiated confession were the type of evidence that could trigger a Schlup inquiry, we would then have to determine whether Carriger is entirely innocent of Shaw’s killing. Proof that others are also guilty, or that Carriger’s role was less central than the original jury may have believed, is not enough. To pass through the Schlup gateway, Carriger must do more than raise doubt in the mind of some imaginary juror. He must identify new evidence which so undermines the state’s ease that each and every reasonable juror would vote to acquit.
Measured against this standard, there is precious little here that helps Carriger. Let’s start with Dunbar’s confession, which the majority seems so taken with. Because Shaw was killed in the course of a robbery, Carriger would be guilty of felony murder even if he did not kill Shaw himself. See State v. Vargas, 127 Ariz. 59, 618 P.2d 229, 230 (Ariz.1980) (en banc). The most likely way for Carriger to prove his innocence would be to show that he had no involvement in the crime whatsoever-that he was somewhere else, minding his own business. Dunbar’s confession helps Carriger only if we think that all twelve imaginary jurors would swallow the preposterous tale that Dunbar and Joyce robbed and killed Robert Shaw together. Yet the record lacks even a sliver of support for this fantasy: There is no physical evidence of her involvement and she has an alibi. Absolutely no one-Carriger himself included-takes this part of Dunbar’s confession seriously.9
*487Joyce’s participation is not a side show-a trivial detail that can be shrugged off. So far as Carriger is concerned, this is the meat of the coconut. Let’s say Dunbar was there, that he participated in the robbery, that he wielded the lethal frying pan. See maj. op. at 475. None of this proves Carriger wasn’t there. The only part of Dunbar’s confession that exonerates Carriger is where he claims it was Joyce (not Carriger) who was his accomplice. And for that part we have only Dunbar’s unadorned words-the fabrication of a proven liar.
But it’s even worse than that for Carriger. Under Schlup we cannot look just at Dunbar’s confession; we must consider all post-trial evidence, including Joyce’s testimony and that of her children.10 According to Joyce, Dunbar and Carriger were in cahoots when they committed the robbery and murder. Joyce had reason to falsely accuse Dunbar, but no motive whatever to falsely accuse Carriger. Larry White, Joyce’s eldest son, reported that when Dunbar let “bits and pieces” of the story slip out over the years, he included Carriger (not Joyce) as the accomplice. Asked directly about Carriger’s involvement, Larry confirmed that “I picked up from Bob [Dunbar] that [Carriger] was involved with the whole thing.” R.T. 7-14-87 at 58. Patience Stevens, the youngest daughter, once overheard Dunbar laughing about the murder: “Bob was laughing, and then he said that-that the guy was begging for Paris [Carriger] not to kill him and telling him that he had family and that, and that he loved them.” R.T. 7-14-87 at 130. Stephanie Stevens, another daughter, confirmed that Joyce was at home in the living room-not out with Dunbar robbing the jewelry store-the afternoon of the murder. It’s hard to imagine anyone-let alone twelve out of twelve reasonable jurors-who would hear the new testimony of Joyce and her children and still swallow Dunbar’s confession.11
Short of proving that Carriger was not involved in the crime, Dunbar’s vacillation might help Carriger in a subtler way by impeaching a central witness for the prosecution. So let’s assume that our imaginary jury would think Dunbar was a liar and disregard his testimony altogether. This wouldn’t help Carriger because the state’s circumstantial case is reasonably strong and stands independent of Dunbar’s trial testimony. Even looking at the case anew, with a stigma of doubt attached to everything that came out of Dunbar’s mouth, reasonable jurors considering the remaining proof could and would still find plenty on which to convict Carriger of felony murder. Carriger knew Dunbar from prison and knew his reputation as an accomplished burglar. Just two days before Shaw’s murder, Dunbar and Carriger made a failed burglary attempt. Carriger moved onto Dunbar’s property the day before the murder. The day of the murder, Carriger called in sick and spent the day with Dunbar. The two of them visited jewelry stores, pawn shops and electronics stores; they also stopped to buy a gun. They walked into Shaw’s jewelry store, where Dunbar commented that it would be an easy place to hit. These certainly appear to be the actions of two veterans planning a robbery.
There is more. Carriger's fingerprint was found on the tape used to bind Shaw’s wrists. *488Forensic post-trial evidence shows that the print could only have gotten onto the tape at the murder scene.- See pp. 490-491 infra. We also have the droplets of blood found on Carriger’s boots. It’s true that they were too small to test, and that they might have come from a source other than Robert Shaw. But it’s not a big leap to conclude that Carri-ger’s boots were splattered with blood at the jewelry store and that Carriger’s feet were in them at the time.
And more still. Carriger has no alibi. Nor did he adequately explain his incriminating actions after the robbery. That evening, after Dunbar supposedly surprised him with the news that he had robbed a jewelry store while wearing Carriger’s boots, Carriger made no effort to distance himself from Dunbar. Instead, he took the two cases-the gun ease and the attache case containing the jewelry-and delivered them to Dunbar’s stepdaughter. Carriger still had the key to the attache case when he was arrested.
A jury confronted with all the evidence at trial, plus Dunbar’s fleeting confession, plus the post-trial testimony of Joyce and her children, plus the forensic evidence putting Carriger at the scene, would not believe every syllable of Dunbar’s confession to the exclusion of everything else, as the majority does here.'12 Far more likely, a rational jury would regard all of Dunbar’s testimony as unworthy of belief. But it probably would accept the explanation of Joyce and her children that Dunbar (who was living with them at the time) pressured them into providing him with an alibi at Carriger’s trial, and credit Joyce’s account that Dunbar and Car-riger committed the crime together. Dunbar’s statements about his participation in the crime-about having bashed in a man’s skull-would serve to implicate Dunbar, but would also count heavily against Carriger as his accomplice. Joyce’s post-trial statement would also tie up other loose ends in the record, such as how Dunbar’s fingerprints got on the jewelry. All told, the case against Carriger-even with Dunbar’s testimony completely omitted-is stronger now than it was at trial. It is inconceivable that twelve rational jurors hearing the ease today would unanimously find Carriger innocent of any and all involvement in Shaw’s murder.
IV
The majority makes another egregious error when it discards the state court’s finding that Dunbar’s confession was not credible. The majority’s methodology will seriously undermine the degree of deference federal courts in this circuit give to state court findings of fact in habeas proceedings. The state court here, it will be recalled, found that Dunbar told the truth when he testified at trial and again when he recanted his confession. - The majority comes to the opposite conclusion, finding Dunbar’s confession credible. - In so doing, it sets aside state court findings of fact which come to us with a presumption of correctness; these findings must be given at least as much deference as findings by our district courts, perhaps a bit more. Indeed, it “would pervert the concept of federalism to conduct a more searching review of findings made in state trial court than we conduct with respect to federal district court findings.” Hernandez v. New York, 500 U.S. 352, 369, 111 S.Ct. 1859, 1871, 114 L.Ed.2d 395 (1991) (citation and internal quotation marks omitted). The majority ignores this guidance and treats the state *489court’s findings of fact with the skepticism usually reserved for findings of Administrative Law Judges and Immigration Judges. See, e.g., Washington v. Garrett, 10 F.3d 1421, 1430-31 (9th Cir.1993) (reversing ALJ); Aguilera-Cota v. INS, 914 F.2d 1375, 1381-82 (9th Cir.1990) (reversing IJ); Turcios v. INS, 821 F.2d 1396, 1400 (9th Cir.1987) (reversing IJ).
The majority first casts aspersions on the credibility findings of Judge Ryan (who handled the state post-conviction proceedings) by noting that he did not also preside at Carriger’s murder trial. See maj. op. at 473-474. But appellate court deference is based on the institutional competence of the factfin-der, not merely his observation of the witnesses. Anderson v. City of Bessemer City, 470 U.S. 564, 574-75, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985). The added concerns of finality and federalism make an even stronger case for deference on collateral review. Nor can the majority take cover in the statutory phrase “as a whole”: “We greatly doubt that Congress, when it used the language ‘fairly supported by the record’ considered ‘as a whole’ intended to authorize broader federal review of state court credibility determinations than are authorized in appeals within the federal system itself.” Marshall v. Lonberger, 459 U.S. 422, 434-35, 103 S.Ct. 843, 851, 74 L.Ed.2d 646 (1983).
In any event, Judge Ryan had plenty of opportunity to observe Dunbar, Joyce, Joyce’s children and other witnesses. Most significantly, of course, he saw Dunbar’s courtroom confession, and six weeks later he saw the recantation. Judge Ryan was therefore practically, not merely institutionally, in a better position to figure out when Dunbar was lying. “Title 28 U.S.C. § 2254(d) gives federal habeas courts no license to redetermine credibility of witnesses whose demean- or has been observed by the state trial court, but not by them.” Marshall, 459 U.S. at 434, 103 S.Ct. at 851.
The majority nitpicks its way through the record and concludes that “each piece of evidence is as consistent with Dunbar’s confession as it is with his trial testimony.” Maj. op. at 473. See also id. at 472 n. 1 (“[T]he inference that Dunbar acted alone is at least equally plausible.”); id. at 474 (“Nothing establishes that the clothes had been worn by Carriger, rather than Dunbar.”) id. at 474 (“The boots are no more consistent with Carriger’s guilt than with Dunbar’s.”); id. at 478 (“other evidence ... points as directly to Dunbar as to Carriger”).13 But if the evidence supports Dunbar’s trial testimony just as well as his confession, how can we reject the state court’s finding that Dunbar told the truth at trial? “Where there'are two permissible views of the evidence, the factfinder’s choice between them cannot be clearly erroneous.” Anderson, 470 U.S. at 574, 105 S.Ct. at 1511. If the majority is holding, as it seems to, that state court findings of fact will only be upheld where the record supports no other scenario, it has stepped directly in the path of a long train of contrary authority. See, e.g., Purkett v. Elem, 514 U.S. 765, 769, 115 S.Ct. 1769, 1771-72, 131 L.Ed.2d 834 (1995); Parke v. Raley, 506 U.S. 20, 35-36, 113 S.Ct. 517, 526-27, 121 L.Ed.2d 391 (1992); Burden v. Zant, 498 U.S. 433, 436-37, 111 S.Ct. 862, 864-65, 112 L.Ed.2d 962 (1991); Demosthenes v. Baal, 495 U.S. 731, 736-37, 110 S.Ct. 2223, 2225-26, 109 L.Ed.2d 762 (1990); *490Wainwright v. Witt, 469 U.S. 412, 434, 105 S.Ct. 844, 857, 83 L.Ed.2d 841 (1985); Maggio v. Fidforck 462 U.S. 111, 118, 103 S.Ct. 2261, 2264-65, 76 L.Ed.2d 794 (1983).
Worse still is the majority’s hopelessly one-sided review of the record. The majority does not look to see whether there is support for the state court’s presumptively correct findings; rather, it chooses only those facts and inferences that undermine those findings. Take the single most damning piece of proof against Carriger: his fingerprint on the adhesive tape used to tie up Shaw’s .wrists. If Carriger’s print got on the tape before the murder, it could only have been on the outside of the reel, which would place it on the first few inches used to tie Shaw’s wrists. Nine feet of tape were used to tie up Shaw,.and most of it was found balled up in a large wad. The only way any preexisting prints could have survived on the tape is if the first six or so inches used to tie up Shaw had been left uncovered and un-smudged when eight and one-half more feet were wrapped around his wrists and wadded up into a ball. Anyone who has used tape will know this is highly unlikely. The usual way of applying tape is. to stick the end piece to the object being bound-be it a duct, a package or a human wrist-and then bring the reel around the object and stick the second layer of tape on top of the first. It would be remarkable to use nine feet of tape to tie up a person’s wrists yet leave the first six-inehes exposed. I have trouble imagining how this might be done.
The jury heard the defense’s theory about the print at trial and, not surprisingly, rejected it. The state has since confirmed this common sense conclusion with expert testimony. A criminalist at the post-conviction proceedings explained that, after cutting the wad of tape into seven pieces so he could unravel it, he fit the pieces together end to end to see where Carriger’s print was located. It was twenty-one and one-half inches from the end of the tape-not on the first few inches. This destroys Carriger’s theory and provides strong proof that he was present and participated in the murder.
The majority agrees that the criminalist’s evidence must be considered, but finds it unreliable and accords it “little weight.” Maj. op. at 474. The majority points to a perceived inconsistency in the expert’s testimony because the criminalist was unable to tell which end of the tape matched the end of thé roll. It is difficult to understand why the majority sees this as inconsistent-either way the fingerprint is not near the end. More remarkable is how quickly my colleagues are willing to dismiss solid evidence that points directly to Carriger’s guilt while blowing out of all proportion scraps of evidence, like hearsay statements in Dunbar’s prison file, that might conceivably point the other way. This is not the sober, dispassionate review of the evidence Schlup calls for; it is a scavenger hunt for anything that might cast doubt on the jury’s verdict. •
The fingerprint evidence is exactly the kind of proof that 'should, be given the greatest weight in making the Schlup inquiry, as the expert witness testified under oath, and Carriger’s lawyer both interviewed him and cross-examined him extensively. And the expert was unequivocal in locating Carriger’s print on the tape:
Q All right. Can you tell us how long the length of tape is approximately?
A Approximately eight feet, nine inches.
Q How many segments are there in that length of tape?
A Twelve.
Q Where is- the segment in Exhibit 43, the latent lift segment of tape-where does that occur from the end of the metal roll of tape contained in the plastic bag in Exhibit 41?
A Approximately 21 and a half inches, ma'am.
Q To what degree of certainty are you certain of this result? Did I ask a bad question? Is there any degree of certainty that you can state that you are confident in your conclusion that you have reassembled the entire roll?
A I am certain that I have reassembled the segment of tape that was torn to make the wad, and I am certain that the latent fingerprint lift segment is in its appropriate place in relation to the length of *491tape and to the piece of tape on the ends of the roll of tape.
MS. FUNKHOUSER: Your Honor, at this time I would move to admit exhibits No. 41, 42, 48 and 54 for purposes of the offer of proof and I have no additional questions for this witness at this time.
THE COURT: Okay. Those exhibits are admitted for the limited purposes of the offer of proof.
R.T. 12-11-87 at 56-57.
The evidence was not excluded for unreliability, as the majority suggests. Rather, it was excluded because Carriger, at the last minute, withdrew his challenge to the fingerprint evidence. Judge Ryan therefore ruled that the testimony from the criminalist was no longer relevant:
The reason I am sustaining the defense’s objection to calling these witnesses to present evidence is because since the defense has abandoned their challenge to the fingerprint testimony, that I don’t think its relevant for you to present evidence, but you want to make an offer of proof to show what the evidence would be if the Court were to allow that.
R.T. 12-11-87 at 6. Following Judge Ryan’s suggestion, the state presented the testimony as an offer of proof, just in case Carriger decided to challénge the fingerprint evidence later on. Carriger has now done so and the testimony of the criminalist must be accorded due weight.
The majority also relies on the fact that only one fingerprint was found on the tape: “Had Carriger left the print while binding the victim, as the state contends, it is difficult to understand how he could have left only one print.” Maj. op. at 474. This is silly. We are not talking here about nine feet of tape laid out flat like an FBI fingerprint card.14 Most of the tape was stuck to itself and rolled up in a big wad, hardly the ideal medium for latent fingerprints: “The development of fingerprints on adhesive tape has always been a challenge for latent fingerprint examiners.” Advances in Fingerprint Technology 88 (Henry C. Lee & R.E. Gaensslen eds., 1994). How many recoverable prints do my colleagues expect to find on such an inhospitable surface? I find it surprising that even a single readable print was retrieved from a wad of adhesive tape cut from the body of a murder victim.
In the end, it matters not whether my view of the evidence or that of the majority prevails. Let’s say the majority’s view is plausible. The contrary view-held by the jury, Judge Ryan, the Arizona Supreme Court, and the district court-surely is plausible as well. What I find alarming is the majority’s methodology, which is to take one view of the evidence, and one view only, ignoring the proof and reasonable inferences pointing the other way. Séction 2254(d) demands far greater deference to state court judges. And not just in degree, but in basic approach: Reviewing the record “as a whole” isn’t.a license to draw every single inference in favor of the petitioner.
We’ve been reversed for this kind of thing before. In Sumner v. Mata, the Supreme Court admonished us that the presumption of correctness could not be east aside through pro forma acknowledgment of section 2254:
Obviously, if the [Ninth Circuit] Court of Appeals ... or any other court of appeals had simply inserted a boilerplate paragraph in its opinion that it had considered the state record as a whole and concluded that the state appellate court’s factual determinations were not fairly supported by the record, this objection to the judgment of the Court of Appeals could not as easily be made. Just as obviously, this would be a frustration of the intent of Congress in enacting § 2254(d).
Sumner v. Mata, 449 U.S. 539, 549, 101 S.Ct. 764, 770, 66 L.Ed.2d 722 (1981). Each un*492derlying fact is entitled to a presumption of correctness unless an exception to 2254(d) applies; federal courts .reviewing state findings on habeas must face up to any disagreement as to each one of these facts. Compare Mata v. Sumner, 696 F.2d 1244 (9th Cir.1983), Mata v. Sumner, 649 F.2d 713 (9th Cir.1981) and Mata v. Sumner, 611 F.2d 754 (9th Cir.1979) with Sumner v. Mata, 455 U.S. 591, 102 S.Ct. 1303, 71 L.Ed.2d 480 (1982) and Sumner v. Mata, 449 U.S. 539, 101 S.Ct. 764, 66 L.Ed.2d 722 (1981). The majority’s willingness to nimbly pick its way through the evidence and come up with its own appellate factfinding bodes ill for the doctrine of deference to state court findings. The presumption of correctness is left in tatters.
V
On the other side of the Schlup gateway, the majority discovers a Brady violation. But my colleagues gloss over a key fact: Before trial Carriger himself wrote his lawyer, Thurman Gay, alerting him to Dunbar’s history as an informant and directing him to get Dunbar’s prison record. Carriger’s letter states in relevant part:
Point # 3
Bobby Dunbar was the informant for the state against John Douglas Troutman for a safe burglary of the National Cash Register Company (about 1968 I think)
Point # 4
Perhaps while we are getting Dunbar’s prison record we should also get a copy of mine.
Attachment to District Court Record on Appeal at 126 (April 18, 1978 letter to Thurman Gay) (emphasis added).
Brady does not require the prosecutor to direct a counter-investigation to destroy its own case. See United States v. McVeigh, 954 F.Supp. 1441, 1449 (D.Colo.1997). Accordingly, a defendant who has specific knowledge of the evidence “withheld” but fails to request it cannot claim a Brady violation. The government need not give notice sua sponte where the accused has knowledge of the documents in question. See Routly v. Singletary, 33 F.3d 1279, 1285 (11th Cir.1994); United States v. Iverson, 648 F.2d 737, 738-39 (D.C.Cir.1981); Ross v. Heyne, 638 F.2d 979, 986 (7th Cir.1980); United States v. Meinster, 619 F.2d 1041, 1045 (4th Cir.1980); United States v. Prior, 546 F.2d 1254, 1259 (5th Cir.1977); Hampton v. United States, 504 F.2d 600, 603 (10th Cir.1974); Wallace v. Hocker, 441 F.2d 219, 220 (9th Cir.1971). The prosecutor in this case breached no duty.
Dunbar’s prison file would have provided further proof that Dunbar was a snitch, a thief and a liar. Carriger alerted Gay to the existence of the file. Gay apparently decided that he had enough to impeach Dunbar. Not a wise choice, perhaps, but neither is it one for which the prosecutor must shoulder the blame under Brady. A prosecutor has no duty to investigate and turn over impeachment evidence the defense already knows about. Carriger’s Brady claim fails.
VI
This is a difficult case. The state relied heavily on the testimony of an unreliable informant. Still, its case at trial was sufficient and nothing that has come to light since the verdict ushers Carriger through the actual innocence gateway. When navigating the murky waters of “fundamental miscarriage of justice,” we ought to stick close to the rules handed down by the Supreme Court and accept the guidance provided. Because I disagree with the majority’s reading and application of Schlup, I dissent.
. Arizona courts share this suspicion:
There is no form of proof so unreliable as recanting testimony. In the popular mind it is often regarded as of great importance. Those experienced in the administration of the criminal law know well its untrustworthy character.
State v. Krum, 183 Ariz. 288, 903 P.2d 596, 602 (Ariz.1995) (en banc), quoting State v. Sims, 99 Ariz. 302, 409 P.2d 17, 22 (Ariz.1965).
. The law indeed accords dying declarations special treatment, at least where the statement concerns the cause of the declarant's impending death, based on the premise that one is not likely to lie so soon before Judgment. See Fed.R.Evid. 804(b)(2).
. Some of the details of Joyce's story are subject to doubt — among other things, she testified that Dunbar took a large frying pan along to rob the jewelry store. (Shaw’s wife explained at trial 'that a small frying pan, identified as the murder weapon, was kept on a hot plate in the store.) Joyce was obviously embellishing to make Dunbar look guilty, thereby providing special motivation for Dunbar to do the same to her.
. This standard, derived from Murray v. Carrier, 477 U.S. 478, 496, 106 S.Ct. 2639, 2649-50, 91 L.Ed.2d 397 (1986), is somewhat less strict ¿an the standards discussed in Herrera v. Collins, 506 U.S. 390, 417, 113 S.Ct. 853, 869, 122 L.Ed.2d 203 (1993) and Sawyer v. Whitley, 505 U.S. 333, 336, 112 S.Ct. 2514, 2517, 120 L.Ed.2d 269 (1992). Before the three-judge panel Carriger’s lawyer abjured Schlup and pressed only a Herrera claim:
Q Let me ask you about the actual innocence claim. Are you making both a Herrera and a Schlup claim?
A No. We are making a Herrera claim. We filed a memorandum accompanying the petition for habeas corpus and it spells out: This is a Herrera claim.
* * s¡< * s}: *
Q ... With respect to the specific question you were just asked'by Judge Kozinski, on pages 15 and 16 of your brief you specifically mention Schlup v. Délo and I guess I’m not quite sure how that plays into the Herrera argument you were just making.
A ... That’s there [because] that’s what the District Court thought we were doing. It didn’t look at the Herrera claim, it went and said, this isn't a claim, standing on its own, of actual innocence; this is a claim that was brought as a gateway under Schlup to resolve the issues of competence and Brady.
Oral Argument (telephone) 11/29/95 10:00 a.m. Relying on this representation, the panel focused its actual innocence analysis on Herrera rather than Schlup. See 95 F.3d 755, 757-61 (9th Cir.1996). Carriger's lawyer resuscitated the Schlup claim at oral argument before the en banc court:
Q You’re talking about actual innocence. Are you talking about that in terms of both Herrera and Schlup, having Schlup act as a gateway?
A That's exactly right.
Oral Argument 3/27/97 10:00 a.m.
. Under Herrera, the petitioner must make a "truly persuasive demonstration of 'actual innocence,’ ” while under Schlup it is sufficient tb show that he is probably innocent. Compare Herrera, 506 U.S. at 417, 113 S.Ct. at 869, with Schlup, 513 U.S. at 327, 115 S.Ct. at 867.
. The source of the majority's confusion may lie in the manner in which the Schlup Court distinguished Herrera. Writing for the Court, Justice Stevens explained that when a petitioner makes a credible showing of innocence-such that a court cannot have confidence in the “result” or "outcome” of the trial-the gateway must open and the case must be considered on the merits. Schlup, 513 U.S. at 316-17, 115 S.Ct. at 861-62. The majority confuses confidence in the fairness of the proceedings with confidence in the result or outcome. Confidence in the outcome of a trial is undermined only if we are persuaded that the petitioner is innocent-even if the original juiy verdict was tainted by constitutional error. This sounds remarkable at first blush, but it really isn't. To avoid swallowing up the doctrine of abuse of the writ, the miscarriage of justice inquiry set forth in Schlup emphasizes actual innocence-the outcome of the trial-rather than the fairness of the trial. Only if we are persuaded that the petitioner is actually innocent can we proceed through the gateway and then consider whether the petitioner received a fair trial.
. The majority seems to adopt the Strickland prejudice standard rather than Schlup ¡Carrier for Ninth Circuit actual innocence claims. Compare maj. op. at 474 n. 4 ("Our task is to determine whether confidence in the actual verdict is undermined.”) with Strickland v. Washington, 466 U.S. 668, 695, 104 S.Ct. 2052, 2068-69, 80 L.Ed.2d 674 (1984) (holding that prejudice requires "a reasonable probability that, absent the errors, the factfinder would have had a reasonable doubt respecting guilt.”) Undermining confidence in the verdict-in the sense that new evidence or flaws in the trial might have led one of the jurors to vote differently-isn't enough to open the gateway. The Schlup standard is higher, and requires the petitioner to prove that he is actually innocent. See Schlup, 513 U.S. at 327, 115 S.Ct. at 867 ("The petitioner thus is required to make a stronger showing than that needed to establish prejudice.”); id. at 332-33, 115 S.Ct. at 869-70 (O'Connor, J., concurring) (fifth vote) (“This standard is higher than-that required for prejudice[.]”).
. Indeed, this was a source of friction between the majority and dissent in Schlup. Justice Stevens defended the standard as manageable; "Finders of fact are often called upon to make predictions about the likely actions of hypothetical 'reasonable' actors. Thus, the application of 'more likely than not' to the habeas court's assessment of the actions of reasonable jurors is neither illogical nor unusual.” Schlup, 513 U.S. at 329 n. 48, 115 S.Ct. at 868 n. 48.
. Dunbar implicated his wife throughout the "confession.” Dunbar claimed that after they arrived at the store, Shaw was still there, so Dunbar bound him up, went over to the cabinets and started taking out pieces of jewelry. Dunbar then presented a vivid picture of the murder:
I started taking things from the cabinets and I heard a muffled kind of thud or gong and I also heard some kind of screams or moans. And' I ran back to the back and I saw lots of blood around Mr. Shaw and his head, and Joyce was standing there saying over and over, "I can’t get caught. I can’t get caught. I got kids.” •
*487R.T. 10-30-87 at 13. Dunbar then admitted that he, too, hit Shaw with the frying pan and strangled him, but only after Joyce had struck the first blows. Id.
In the context of Joyce’s testimony and the other evidence that Carriger and Dunbar committed the crime together, Dunbar’s confession can be convincingly read as the very real recollection of the crime if Carriger is substituted for Joyce as striking the first blows with the skillet.
. Carriger also identifies hearsay evidence of Dunbar boasting to iriends about how he set Carriger up. See maj. op. at 472. It’s not clear if any of this is reliable enough to form the basis of a Schlup claim, but it doesn’t matter because it all suffers from the same fatal flaw as Dunbar's confession: It fails to show that Carriger is actually innocent of murder, and it too suggests only that Dunbar committed the murder with Carri-ger and set Carriger up after the fact to take the fall by himself.
. It’s even harder for me to imagine how the majority can review this record and still say that "[tjhere is no evidence that Dunbar and Carriger acted together to commit the murder.” Maj. op. at 472 n. 1. Is the testimony of Joyce and her children "no evidence”? This testimony alone is enough to discredit Dunbar’s confession and seal Carriger's fate.
. Dunbar’s explanations during this confession are the road map the majority uses to explain away all physical evidence of Carriger’s involvement. See maj. op. at 473 ("Dunbar explained that he had taken the adhesive tape ..."); id. at 474 ("Dunbar testified that he gave Carriger the attache case to give to Jackie White to hold;”); id. at 474 ("Dunbar testified in his confession that the clothes he led the police to were the clothes he, Dunbar, had worn during the robbery”); id. at 474 ("[W]hen he confessed, Dunbar stated he planted Carriger’s jeans in the canal”); id. at 474 ("In his confession, Dunbar testified that he discarded the clothes ... ”); id. at 478 ("Dunbar, without immunity, confessed under oath to committing the murder”); id. at 478 ("Dunbar’s confession explained why Carri-ger’s fingerprint was on the tape, although Carri-ger was not there”).
One would hardly believe it’s the selfsame Dunbar the majority describes as a “known habitual liar,” id. at 465, a "habitual liar,” id. at 466, a "less than reliable” informant, id. at 470, "well known ... to be a liar,” id. at 470, "manipulative” and "sociopathic,” id. at 471, 481, "lying,” id. at 475, "vague and evasive,” id. at 475, and saddled by a "long history of lying to police," id. at 479, 481.
. The majority draws each inference in favor of Carriger. It begins by discarding the physical evidence, finding that: Carriger could not have left only one fingerprint on the tape while binding Shaw's hands, maj. op. at 473; Carriger's having the key to the attache case meant nothing, id. at 474; Dunbar wore Carriger's clothes during the robbery, id. at 474; Dunbar planted Car-riger’s jeans in the canal, id. at 474; Carriger's fingerprints on the gun case meant nothing, id.- at 474; Dunbar wore Carriger's boots, id. at 474; and Carriger would have left recoverable fingerprints on the jewelry had he not worn gloves, id. at 474. The majority then moves on to even more speculative matters, finding that: Dunbar could not have known the details of the crime unless he participated, id. at 475; Dunbar was not shown diagrams of the store; id. at 475; Dunbar's lawyer was not interested in publicity, id. at 475; Dunbar was vague and evasive when recanting his confession, id. at 475. None of these findings had been made before in the twenty-year history of this case, and each is used to support the majority's most speculative (and contradictory) findings of all: that Dunbar was telling the truth when he confessed, id. at 475, and that he acted alone, id. at 472 n. 1. The role of the factfinders in our criminal and habeas systems demands more respect than this.
. Even those little cards aren't foolproof. Working under ideal conditions-a willing subject, a flat card clasped firmly into place, fingers drenched in printer's ink-it often takes experts several tries to produce a usable set of prints. Under less ideal conditions, obtaining a usable print is very chancy. It depends on several factors: The kind of surface on which the print might be left, how oily or dry the individual's skin is, and whether anything has smudged the print. See Advances in Fingerprint Technology 60-63 (Henry C. Lee & R.E. Gaensslen eds., 1994).